Lifelike eyes that follow you. Built with Raspberry Pi.
A real-time rendered parallax eye system that tracks faces, reacts to sound, and responds to depth — designed for Burning Man, Halloween, art installations, or anywhere you want something watching.
Live Demo · Build Guide · Hardware List
Two round HDMI displays show procedurally rendered 3D eyes that:
- Track faces — OpenCV DNN + MediaPipe detect faces and follow them with parallax layers
- React to depth — pupil dilates dramatically as you get closer (0.6x → 1.6x range)
- Follow motion — hands, bodies, any movement — not just faces
- Pulse to music — stereo microphone detects bass beats, pupil throbs in sync
- Startle at sounds — loud claps trigger a blink + pupil spike
- Look toward noise — stereo audio direction makes the eye turn toward sounds
- Blink naturally — randomized blink intervals with double-blink chance
The eye is rendered as three parallax layers (sclera, iris, pupil) that move independently — the pupil tracks furthest, the sclera barely moves. This creates a convincing depth illusion on flat screens.
Detection runs on a background thread at 15 fps. Rendering runs at 60 fps with smooth interpolation between detection frames.
| Part | Price | Why |
|---|---|---|
| Raspberry Pi 5 (8GB) | $80 | Quad-core ARM, dual HDMI, PCIe for AI HAT |
| 2x Round HDMI Display (1080x1080) | ~$50 each | The "eyeballs" |
| Pi Camera Module 3 (or USB webcam) | $25-35 | Face/motion detection |
| Pi AI HAT+ 13 TOPS (optional) | ~$77 | NPU acceleration for smoother tracking |
| Pi AI Camera IMX500 (optional) | $70 | On-sensor inference, zero CPU cost |
| USB-C Power Bank (20,000mAh+) | ~$30 | Portable power (~6 hours runtime) |
Total minimal build: ~$240 · Full build with AI acceleration: ~$380
Flash Pi OS Desktop (Bookworm 64-bit) with Raspberry Pi Imager. Enable SSH, set username pi, password raspberry, hostname raspieyes.
- Plug both round displays into HDMI ports
- Connect camera (CSI ribbon cable or USB)
- Power on
scp -r raspieyes/ pi@raspieyes.local:~/
ssh pi@raspieyes.local 'bash ~/raspieyes/setup.sh'Edit config.txt:
RENDER_MODE=parallax # Real-time rendered eyes (or "video" for video loop)
TRACKING=yes # Enable camera tracking
EYE_COLOR=blue # blue, green, brown, amber, gray
USB_WEBCAM=no # Set to "yes" if using USB webcam instead of Pi Camerasudo rebootThe eyes start automatically on boot. Walk in front of the camera!
All settings in config.txt:
# Rendering
RENDER_MODE=parallax # "parallax" (real-time) or "video" (loop mp4s)
EYE_COLOR=blue # Iris color
# Tracking
TRACKING=yes # Enable camera
DETECTION_MODE=motion # "motion" or "face"
USB_WEBCAM=no # USB webcam instead of Pi Camera
# Tuning (uncomment to customize)
# MIN_CONTOUR_AREA=300 # Motion sensitivity
# PRESENCE_TIMEOUT=15 # Seconds to hold position after losing track
# SCLERA_PARALLAX=0.08 # Parallax intensity per layer
# IRIS_PARALLAX=0.6
# PUPIL_PARALLAX=0.95
# MAX_OFFSET=350 # Max pixel displacementYou can test the eye renderer on your Mac before deploying to Pi:
# Mouse tracking (no camera needed)
python3 eye_renderer.py --mouse --windowed --single-eye
# Webcam tracking (run from Terminal.app for camera access)
python3 eye_renderer.py --test-webcam --windowed --single-eye
# Dual eyes
python3 eye_renderer.py --mouse --windowedeye_renderer.py — Main renderer: parallax eye, animation, audio, 60fps loop
eye_tracker.py — Detection functions: face (DNN/Haar), motion (MOG2), camera setup
tracker.py — Centroid tracker with Kalman filtering
play.sh — Autostart script: screen setup, renderer launch, watchdog
config.txt — Runtime configuration
setup.sh — One-time Pi setup: deps, autostart service
sudo systemctl status raspieyes # Check status
sudo systemctl restart raspieyes # Restart
sudo systemctl stop raspieyes # Stop
journalctl -u raspieyes -f # Live logsMIT
Built by Alejandro for Burning Man 2026.